Refined Composite Multiscale Range Entropy and Pairwise Feature Proximity-Based Fault Detection Method of Rotating Machinery

Purpose: Rotating machinery is widely applied to modern industrial equipment as a typical transmission component. Accurate and fast detection of the fault states of rotating machinery is vital to guarantee safe production. Vibration signals can indicate the fault states, but these signals are difficult to use directly. This work is dedicated to the study of mining important feature information from signals and classifying the features to enable the detection of rotating machinery faults. Methods: First, a complexity measurement method called refined composite multiscale range entropy is proposed to describe the working condition of rotating machinery. Further, a fault detection scheme based on the proposed entropy, pairwise feature proximity, and kernel extreme learning machine optimized by the von Neumann whale optimization algorithm is presented. Specifically, high-quality fault features are derived with the entropy method from the vibration signals to form a feature matrix. Then, the pairwise feature proximity algorithm is adopted to obtain low dimensional features with high resolution. At last, the critical parameters of the kernel extreme learning machine are optimized by the von Neumann whale optimization algorithm to get the best classification performance for fault detection. Results: The developed fault detection scheme is assessed with three rotating machinery fault datasets. The experimental results indicate that the scheme can achieve the best fault detection effect compared with other fault detection schemes. Conclusion: The proposed refined composite multiscale range entropy can effectively extract fault information from vibration signals, and the proposed fault detection method can accurately detect different types of rotating machinery faults.